Head-Free, Remote Eye-Gaze Detection System Based on Pupil-Corneal Reflection Method With Easy Calibration Using Two Stereo-Calibrated Video Cameras

We have developed a pupil-corneal reflection method-based gaze detection system, which allows large head movements and achieves easy gaze calibration. This system contains two optical systems consisting of components such as a camera and a near-infrared light source attached to the camera. The light source has two concentric LED rings with different wavelengths. The inner and outer rings generate bright and dark pupil images, respectively. The pupils are detected from a difference image created by subtracting the bright and dark pupil images. The light source also generates the corneal reflection. The 3-D coordinates of the pupils are determined by the stereo matching method using two optical systems. The vector from the corneal reflection center to the pupil center in the camera image is determined as r. The angle between the line of sight and the line passing through the pupil center and the camera (light source) is denoted as theta. The relationship theta = k vertical bar r vertical bar is assumed, where k is a constant. The theory implies that head movement of the user is allowed and facilitates the gaze calibration procedure. In the automatic calibration method, k is automatically determined while the user looks around on the PC screen without fixating on any specific calibration target. In the one-point calibration method, the user is asked to fixate on one calibration target at the PC screen in order to correct the difference between the optical and visual axes. In the two-point calibration method, in order to correct the nonlinear relationship between theta and vertical bar r vertical bar, the user is asked to fixate on two targets. The experimental results show that the three proposed calibration methods improve the precision of gaze detection step by step. In addition, the average gaze error in the visual angle is less than 1 degrees for the seven head positions of the user.